Extraordinary claims require verification. MaineCoon's published metrics — 47.5 FPS, sub-3-second first frame, 0.934 SocialVideo Bench score — are impressive. Here's how to check them yourself.
Step 1: Live streaming test
Visit the official Experience Platform at mainecoon.tech/experience-platform. Input a character description and dialogue prompt. Measure time from submission to first visible frame — expect under 3 seconds. Observe whether output streams continuously or appears all at once.
Step 2: Audio-visual sync check
During streaming, focus on lip movement relative to speech audio. Because MaineCoon generates both jointly, sync should be tight without the lag typical of post-hoc dubbing. Try prompts requiring varied speech pacing — pauses, emphasis, emotional shifts.
Step 3: Mid-stream interaction
While generation is active, inject a new instruction — change the character's emotion, ask a question, or shift the topic. Verify the stream adapts without resetting. This tests the interactive control and Buffer Controller behavior.
Step 4: Duration stability
Let generation run for several minutes. Watch for character distortion, scene drift, or quality degradation over time. Official demos show stable 2+ minute streams; the architecture targets 10+ minutes.
Step 5: Review benchmark data
Read the arXiv technical report (2606.17800) for SocialVideo Bench methodology and results. Check Hugging Face and GitHub for model weights, inference code, and reproducibility details. Compare published FPS numbers against your own hardware if self-deploying.